Research On Soybean Futures Price Forecast Based On Complex Network And Neural Network

Originating from China,soybean has a long-standing history of cultivation.For millennia,China has been the world’s preeminent soybean producer.In 1954,China’s soybean production was surpassed by the United States,followed by Brazil and Argentina,and now it is the fourth soybean producer in the world.From 1 million tons imported in1995 to 96.52 million tons in 2021,China’s soybean imports are increasing year by year.China has also changed from a net exporter of soybeans to a net importer of soybeans.China’s dependence on soybeans exceeds 80%.Therefore,international trade policies,international situations,import prices of soybeans and transportation costs will have an important impact on the price of soybeans,and will also lead to the instability of soybean prices.However,China lacks the corresponding voice and pricing power for soybean prices,resulting in a huge impact on China’s domestic soybean production and the production and operation of soybean related processing enterprises due to the fluctuation of international soybean prices.In order to ensure the stable and healthy development of the soybean market and soybean related enterprises,and prevent the adverse effects of market risks and soybean prices,the price discovery function of soybean futures market needs to be fully exploited.Therefore,the accurate prediction of soybean futures price is helpful to the production and operation decision-making of enterprises and the formulation of national economic policies.This thesis takes CBOT soybean futures price data as the research object,and proposes a new hybrid learning method based on complex network and machine learning.Firstly,the corresponding soybean price fluctuation network is established based on the coarsening method,and the soybean futures price network is analyzed from the four perspectives of node strength,clustering coefficient,average path length and intermediary centrality.The results show that the soybean futures price network has the characteristics of node strength and intermediary centrality with power law distribution,small clustering coefficient,short average path length,etc.Then,the topological structure of the network is used to extract the volatility characteristics of soybean futures prices,so as to reconstruct the time series of soybean futures prices.Secondly,because the soybean reconstructed prices have the complex characteristics of nonlinear,non-stationary and high volatility,traditional prediction models and a single machine learning model are difficult to predict them with high accuracy,so in the reconstruction data of predicting soybean futures prices,In this thesis we first decompose the reconstructed data using complete set empirical mode decomposition with adaptive white noise(ICEEMDAN)to obtain sub sequences of different frequencies;Then,Levy’s improved Sparrow Search Algorithm(ISSA)is used to optimize the weights and thresholds of the Extreme Learning Machine(ELM),predict the decomposed subsequences,and finally integrate the predicted results to obtain the final prediction results.At the same time,compared with other benchmark models,the SPN-ICEEMDAN-ISSA-ELM hybrid learning method proposed in this thesis has high prediction accuracy and stability.The main contributions of this thesis are as follows:(1)Reconstruct the soybean futures price data set using complex networks,effectively extract the volatility characteristics of soybean futures prices,and improve the prediction accuracy of soybean futures prices.(2)The decomposition algorithm is used to decompose the data set of soybean futures,which can effectively reduce the prediction difficulty.(3)Levy flight strategy is used to improve the SSA algorithm,and the improved SSA algorithm is used to optimize the ELM model,which improves the prediction accuracy of the ELM model.(4)The SPN-ICEEMDAN-ISSA-ELM hybrid learning method proposed in this thesis can make accurate and stable predictions of soybean futures prices.